On Tue, May 16, 2023 at 05:52:03PM +0100, Mark Rutland wrote:
> Hi Paul,
>
> On Sat, May 13, 2023 at 06:14:36PM -0700, Paul E. McKenney wrote:
> > On Sun, May 14, 2023 at 08:58:00AM +0900, Akira Yokosawa wrote:
> > > Running "./scripts/kernel-doc -none include/linux/atomic/atomic-arch-fallback.h"
> > > on the tag emits a lot of warnings.
> > >
> > > Looks like there are kernel-doc comments who don't have a corresponding
> > > function signature next to them.
> > >
> > > /**
> > > * function_name() - Brief description of function.
> > > * @arg1: Describe the first argument.
> > > * @arg2: Describe the second argument.
> > > * One can provide multiple line descriptions
> > > * for arguments.
> > > *
> > > * A longer description, with more discussion of the function function_name()
> > > * that might be useful to those using or modifying it. Begins with an
> > > * empty comment line, and may include additional embedded empty
> > > * comment lines.
> > > */
> > > int function_name(int arg1, int arg2) <---
> > >
> > > Note that the kernel-doc script ignores #ifdef -- #else.
> >
> > Me, I was thinking in terms of making this diagnostic ignore
> > include/linux/atomic/atomic-arch-fallback.h. ;-)
> >
> > The actual definitions are off in architecture-specific files, and
> > the kernel-doc headers could be left there. But there are benefits to
> > automatically generating all of them.
> >
> > Another approach might be to put a "it is OK for the definition to
> > be elsewhere" comment following those kernel-doc headers.
> >
> > Any other ways to make this work?
>
> I've spent the last day or so playing with this, and I think we can do this by
> relegating the arch_atomic*() functions to an implementation detail (and not
> documenting those with kerneldoc), and having a raw_atomic*() layer where we
> flesh out the API, where each can have a mandatory function definition as
> below:
>
> /**
> * raw_atomic_fetch_inc_release() - does a thing atomically
> *
> * TODO: fill this in
> *
> * This is a version of atomic_fetch_inc_release() which is safe to use in
> * noinstr code. Unless instrumentation needs to be avoided,
> * atomic_fetch_inc_release() should be used in preference.
> */
> static __always_inline int
> raw_atomic_fetch_inc_release(atomic_t *v)
> {
> #if defined(arch_atomic_fetch_inc_release)
> return arch_atomic_fetch_inc_release(v)
> #elif defined(arch_atomic_fetch_inc_relaxed)
> __atomic_release_fence();
> return arch_atomic_fetch_inc_relaxed(v);
> #elif defined(arch_atomic_fetch_inc)
> return arch_atomic_fetch_inc(v)
> #else
> return raw_atomic_fetch_add_release(1, v);
> #endif
> }
>
> ... and likewise we can add comments for the regular instrumented atomics.
I do like that approach! It should be easy to adapt the kernel-doc
scripting to this.
> I've pushed out the WIP patches to my atomics/fallback-rework branch; if you're
> happy to give me another day or two I can get a bit further.
An RCU issue currently has me by the ankle, so I am quite happy to give
you a day or two. ;-)
Just FYI, I will be in the air this coming Friday, your time.
> > For me, the option of making this
> > diagnostic ignore include/linux/atomic/atomic-arch-fallback.h has
> > considerable attraction.
>
> It's certainly appealing...
But I do like your approach of simply always having the function prototype
available. ;-)
Thanx, Paul
